DESCRIPTION (adapted from the Abstract): The rapid and continuing proliferation of on-line biomedical information sources has necessitated the creation of software tools for automatically navigating these networks, identifying relevant documents, and extracting the desired information from them. Designing, implementing, and evaluating such a software system is the primary objective of this project. More specifically, a central aim is to create a system that can be personalized to the particular and changing interests of individual biomedical scientists. Machine leaming methods play a central role in our software assistant allowing the system to adapt to individual users. A flexible language with which scientists can communicate their interests to the machine leaming algorithms will be developed. Machine leaming methods typically have the weakness of requiring user-provided training examples, but creating such examples is usually a burdensome task. Hence, another central aim is to greatly reduce the need for human-labeled training data by creating techniques that obtain these examples by indirect methods. The system being developed will be field-tested in the laboratories of several biomedical researchers.